Koilpillai CR Keynote

Electrical Engineering
IIT Madras
Signal Processing Aspects in Cognitive Radio

- An Overview
R. David Koilpillai
Department of Electrical Engineering
Indian Institute of Technology Madras
International Workshop on Cognitive Radio (IWCR 2010)

IIIT Bangalore – December 15, 2010
Koilpillai / Dec 2010 / Cognitive Radio 1

IWCR 2010 Cognitive Radio Keynote IITM Proprietary Information
Outline of Presentation IIT Madras
Wireless today – SDR, Wireless Broadband, …

Cognitive Radio – a paradigm shift
Different approaches Spectral sensing
Multi-Carrier Techniques
802.22 – A CR-based standard
Major CR initiatives
Cognitive Radio Test-beds
Cognitive Applications in wireless
Focus on the signal processing aspects in all topics

Wireless in the Media … IIT Madras
Cellular
India crosses 635M subs
Growth of broadband wireless
– HSPA, EV-DO, WiMAX
GSM subscribers crossed 4.7 Billion ...
Convergence, Quad play …
– Voice, Voice, Data, Video, Mobility
– Telephony, Internet, TV, Cellular
Spectrum auctions April - June 2010
– Introduction of 3G - 4G Technology
Broadband Wireless Access
ITU Activity – IMT Advanced (4G)
UMTS + LTE > 535M
WiMAX deployments: 593
Cellular Evolution Timeline Electrical Engineering
IIT Madras
1G (AMPS, NMT, TACS, …) 1981

Analogue voice transmission
2G (GSM, IS-54, PDC, IS-95) 1991 - 95

Digital cellular
Digital voice, low-speed circuit data (9.6 Kbps), SMS
2.5G (GPRS, cdmaOne) 1999 - 00

Introduction of packet data
Improved voice, medium speed CS and PS data (~100 Kbps), enhanced SMS
3G (WCDMA, EDGE, cdma2000) 2002 - 03

IMT-2000 requirements, Improved voice, high speed PS data (384Kbps - 2 Mbps)
Improved spectral efficiency and capacity, Multimedia applications
3.5G (HSPA, 1xEV, IEEE 802.16e) 2003 - present

High speed packet data (2-14 Mbps)
4G (LTE, IEEE 802.16m, …) 2010+

Evolution beyond 3G IIT Madras
Rel. 7
Rel. 6
GSM Rel. 5
GPRS WCDMA (HSDPA) LTE-Adv
LTE
1xEV-DV Super 3G = 3.9G
cdmaOne cdma2000
1xEV-DO
IEEE IEEE
802.16 d/e 802.16 m
3.5G/4G - Focus on high data rates, spectral efficiency

Family of Networks IIT Madras
Hierarchy of terrestrial networks

India has all scenarios
Different types of wireless
networks
Wide range of data rates, range
Significant developments in
WAN / MAN
Range of environments
– Dense urban
z Inter-BS distance < 500m
– Sparse rural
Unlicensed devices
Ref: Cordeiro et al., “IEEE 802.22: The First Worldwide
Wireless Standard based on Cognitive Radio,” IEEE, 2005

Broadband Usage Electrical Engineering
IIT Madras
Ref: Kori (Alcatel Lucent) –

WiMAX Overview, Bangalore,
Jan 2009
Compelling case for

broadband wireless
access (BWA) in India
Broadband, wireless –
Quad play
Voice, Data, Video,
Mobility

IMT Advanced IIT Madras
ITU initiative (started 06/2003)

– Development of systems beyond IMT-2000 (open and global standards)
Enhanced IMT-2000 systems should support
– Evolution of new applications, products, and services
Aspects: Enhanced mobile access, Seamless networking.
Data rate requirements:
– 100 Mbits/s for high mobility
– 1 Gbits/s for low mobility (nomadic/local wireless) access
Timeline:
– Initial proposals for IMT-Advanced submitted - 2009
– Standardization 2009 - 2011
Current global standards
– IEEE 802.16m, 3GPP LTE Advanced
– All based on Orthog Freq. Division Multiple Access (OFDMA) & MIMO
India Spectrum Scenario IIT Madras
Spectrum auctions (3G+BWA) held during April-June 2010

– 2x20 MHz TDD (in 2.3 GHz and 2.5 GHz), 22 Service Areas
– In 3G core band, two-four (5+5) MHz FDD slots, 22 Service Areas
– Some differences based on geographic location
Auction method
– Start from “reserve price”
– Bidders will be asked to increase in 10% steps
– Bidding continues until number of bidders left = no. of available frequency slots
Total revenue from auctions
– 3G Spectrum = Rs. 67,719 crores
– BWA Spectrum = Rs 38,543 crores
A deviation from the earlier model (for 2G spectrum)
– Low license fee + revenue sharing model
– Very successful model
Market Case study IIT Madras
Dense Urban (Case: Mumbai)

– 70% of 16M people
z In area of 600 sq Km
– ~3733 households per sq km

z Assuming 5 per household
– ~ 50% wireless internet subscribers

– ~ 1866 wireless internet/sq km
– cell radius = 0.75 km
– ~ 3300 subscribers/cell
– Assuming 5 competitive operators in each area =>
660 subscribers/operator/cell
Typical scenarios evaluated by Indian operators

Participating Operators
– Tata Teleservices, BSNL, Airtel, Reliance, Hutch, IDEA Cellular, Aircel, VSNL, MTNL

IIT Madras
Software Defined Radio (SDR)

Radio Functionality Evolution Electrical Engineering
IIT Madras
Source: Prasad et al. IEEE Comm Magazine, April 2008

Software Defined Radio (SDR) IIT Madras
J. Mitola, “The software radio architecture”

IEEE Communications Magazine, May 1995

DSP view of Software Radio IIT Madras
Software radio approach (cost effective and computationally efficient)

z Extensive use of DSP techniques (including A/D)
z Multirate DSP and filtering
Current situation for wireless systems
– Multiple standards and MIMO (cellular 3/3.5/4G, WiMAX, WLAN, …
z Systems with different bandwidth, channel spacing, symbol rates
– Narrowband systems – requiring large number of transceivers

Conventional approach (multi-channel receiver)
RF stage IF stage Baseband

Bandpass filter A/D
Filter, mixer Filter, (I,Q) mixer processingr
SDR approach (multi-channel receiver)
RF stage IF stage High speed DSP-based Baseband

Bandpass filter
Filter, mixer (I,Q) mixer A/D Channeliser processing

A/D Challenges in Software Radio Electrical Engineering
IIT Madras
Magnitude Response Strong signal
Weak signal
Quantization noise
Frequency →
Very stringent A/D requirements

– Speed, dynamic range
– Dynamic range determined by max. variation bet. strongest and weakest signal
z No clipping of strongest signal
z Enough SNR to detect weakest signal
SDR ⇒ Flexible and efficient receiver and transmitter architectures

SDR poses system design challenges – clocks, sampling rate, filtering, …
Very stringent real-time requirements
COTS SDR Architecture IIT Madras
Commercial product
– Vanu Inc
Multistandard
– GSM / GPRS / EDGE
– Cdma / EV-DO
Flexibility
Scaleability
Cost-effectiveness
Ref: www.vanu.com
Vanu SDR Architecture IIT Madras
Ref: www.vanu.com

SDR Summary IIT Madras
Many technical challenges have been solved

SDR – now commercially viable and attractive
Drivers for SDR
– Advances in processors, DSPs, FPGAs, …
– High speed, high-resolution A/D, …
– Multi-standard support, MIMO capability, …
– Efficient software tools and structures
SDR: A flexible platform
– New technology development such as 4G systems
– Technology migration
Focus on basestations and not user equipment (UE)
Numerous national and international initiatives
– Multiple SDR test beds
– Open-source material available
SDR Forum – an active group
The next step in SDR → Migration towards Cognitive Radio …
IIT Madras
SDR → Cognitive Radio

Cognitive Radio Motivation IIT Madras
Increasing demand for radio spectrum

– Broadband wireless demand is rapidly growing
Current approach to spectrum allocation
– Fixed allocation to licensed users
Existing scenario
– Under-utilization of spectrum
– Spatial and temporal “spectral holes” exist
Innovative approach to improve spectrum utilization
– Cognitive Radio
Initiated by FCC – regarding secondary usage of spectrum
Cognitive Radio techniques – much broader than DSA

– A radio that is aware of its surroundings and adapts intelligently
– Reed et al.

Cognitive Radio IIT Madras
Increasing demand for spectrum

Existing scenario
– Under-utilization of spectrum
Innovative approach to improve spectrum
utilization
– Cognitive Radio
Ref: M.A.McHenry, “NSF Spectrum Occupancy Measurements Project Summary,” August 2005
Ghasemi and Sousa, IEEE Communications Magazine, April 2008
CR Scenario IIT Madras
CR: Opportunistic Unlicensed Access

– To temporarily unused frequency bands (across the entire licensed radio spectrum)
A means to increase efficiency of spectrum usage
Stringent safeguards required
– On-going licensed operations should not be compromised
Spectrum sensing based access
– White spaces – primary user absent, and free of RF interferers
– Gray spaces – primary user absent but partially occupied by interferers
– Black spaces – primary user present
Main functionality of Cognitive Radios
– Ability to reliably identify unused frequency bands
– Sensing must be reliable and autonomous
Radically different paradigm
z Secondary (unlicensed) users - Opportunistic use of unused licensed bands
z Inceased utilization of radio spectrum
TV Bands IIT Madras
White spaces exist in TV bands
Channels 14-69 in 470-806 MHz
– 6 MHz per channel (7 MHz or 8 MHz based on country)
Excellent propagation characteristics in this frequency band
Predictable spatio-temporal usage of TV channels
IEEE 802.22
– An air-interface (PHY & MAC)
– Opportunistic secondary access to TV spectrum
– Safeguards to protect primary user
z From secondary user interference
– First commercial application of Cognitive Radio
– Dynamic Spectrum Access (DSA)
Consider different options (to facilitate secondary users)
Focus on reliable and robust spectrum sensing
– @ -116 dBm

IIT Madras
SDR → Cognitive Radio
Cognitive Radio =
SDR + Sense + Learn + Adapt + Use
IIT Madras
Spectrum Sensing

Methods of Spectrum Sensing IIT Madras
Energy Detector
Correlation-based detector
Cyclostationarity-based detector
Hybrid Detector
Filter bank Method – Multi-taper Method (MTM)
Performance of spectrum sensing
Sensing Criteria (Regulatory aspects)
– Sensing Period
– Detection Sensitivity

Aspects of Spectrum Sensing IIT Madras
Time-varying channel
Lack of apriori information
– SNR level, interference, …
Signal blockage (shadowing, hidden-node…)
Primary signal transition ON ↔ OFF
Single shot detection vs. sequential detection
Interference due to other CR users
Decentralized vs centralized approach
– Cooperative sensing

Spectrum Sensing IIT Madras
Optimum receiver
– If structure of primary signal known
– Optimum (in AWGN): Matched Filter (MF) followed by Threshold
– Not practical for large # of primary users
z Need for coherent detector for each transmitted signal
Alternative – Energy Detector

– Measures energy of signal in primary band
– Compare with properly set threshold
– Requires longer sensing time to achieve desired level of performance
– Low computational complexity
ED - An attractive candidate for Cognitive Radio
Drawbacks of ED
Cannot discriminate between sources of input energy (signal vs. noise)
–
– Uncertainty of noise floor will degrade performance - Especially at low SNR
ED can be effectively combined with more robust detectors – “Hybrid Detectors”

Spectral Sensing IIT Madras
Binary hypothesis testing problem
H 0 : y [n] = w [n] ⇒ Primary User absent

H1 : y [n] = x [n] + w [n] ⇒ Primary User present
n = 0 ,1, L (N-1 ) ( N − sample observatio n window of received signal)
x[n] = transmitte d signal
w[n] = noise (zero − mean AWGN with variance σ w2
y[n] = transmitte d signal
Decision statistic (Energy detector)
N −1
1
Δ= ∑ Δ≥γ ⇒ H1
2
y [ n] and
N n =0
Δ<γ ⇒ H0
Signal absent, Δ is Central Chi-Square Variable with N degrees of freedom
When signal present, non-Central Chi-Square Variable
Energy Detector IIT Madras
Decision statistic
N −1
1
Δ= ∑ Δ≥γ ⇒ H1
2
y[n] and
N n =0
Δ<γ ⇒ H0
If N large, invoke CLT

⎛ 2 σ2
Δ ~ Normal ⎜ σ w , w
( )
2
⎞
⎟ for H 0
⎜ N ⎟
⎝ ⎠
⎛ 2
Δ ~ Normal ⎜ (σ x + σ w ),
2 σ 2
x + σ 2
w ( )
2
⎞
⎟ for H1
⎜ N ⎟
⎝ ⎠
⎛
Pmissed-detection = Q⎜⎜ 2
N
σ 2
+ σ (
w −γ
2
)⎞⎟⎟
⎝σ x +σw
2 x
⎠
⎛
2⎜
Threshold γ = σ w 1 +
Q −1
(Pfa ) ⎞
⎟
⎜ N ⎟⎠
⎝
Energy Detector Performance IIT Madras
Primary user signal → GSM signal

Signal bandwidth 200 KHz
One time slot ⇒ 142 data symbols (incl. 26-bit midamble)
+ 14.25 symbols (Tail + Guard symbols)
Assume one sample / symbol
8 slots ⇒ 1136 samples
Noise power = -116 dBm
– Using noise figure = 7 dB for receiver
Performance plots for Energy Detector

– Good performance at very low SNRs ~ -8 dB
ED is a strong candidate for first stage of Hybrid detector

Spectral Sensing Performance (2) IIT Madras
AWGN, Effect of sensing Period Performance in fading
Robustness of energy detector enhanced if longer sensing period is used

Performance in fading is poorer than in AWGN (as expected)
– Noise uncertainty causes major degradation in performance
Energy detector not suited as a stand-alone detector
Correlation Detector IIT Madras
Well-suited for signals like GSM

– Training sequence in every burst
– Normal burst – 26 symbols
– Sync burst – 64 symbols
N −1
Rx x m =
( )
1
∑ x (n ) x ∗
(n − m )
N n =0
Correlation provides processing gain

– Against noise and interference
Better performance than ED
Computational complexity is higher
– Needs to be done at oversampled rate
Used if primary user signal structure known

Cyclostationary Detector IIT Madras
Cyclostationary Property – Digital modulated signals

Let x (t ) be a zero-mean complex signal
[ ]
–
– The autocorrelation function is given by Rx x (t ,τ ) = E x(t ) x ∗ (t + τ )

– If Rx x (t ,τ ) is periodic in t ⇒ CYCLOSTATIONARY
– Many of the communications signals have cyclostationarity property
– Fourier Series representation Rx x (t ,τ ) = ∑ Rα α (τ ) e
xx
j 2π α t
mult of
– Rαx x (τ ) - cyclic autocorrelation function

z Discrete in α, Continuous in τ
Rα ∗ (τ ) - conjg cyclic autocorrelation and FT S α ∗ ( f ) conjg spectral corr density funcn
–
for which R ∗ (τ ) and S ∗ ( f ) are non-zero depends on

xx xx
The values of α
α α

xx xx
– Symbol rate
Cyclic signature of signal can be exploited to detect signal
– Even in presence of noise and interference

Cyclostationary Property of GMSK IIT Madras
Cyclic Autocorrelation Spectral correlation
Focus on GMSK signal (BT=0.3)

fs
Discrete cyclic autocorrelation with peak at α = ±
2
Pioneering work by Gardner

Cyclostationary Detector IIT Madras
AWGN with 1 & 2 alpha Performance in fading
Exploiting multiple cyclic frequencies improves performance

Cyclostationary feature detector is sensitive to frequency-selective fading
Detector is also sensitive to frequency offset
Cyclostationary feature detector is an attractive method for overall performance

Low-Complexity Hybrid Detector for GMSK IIT Madras
Performance in fading
A robust detector for GSM signals (GMSK)

Energy detector has low computational complexity
– Practical scenario of 1 dB noise uncertainty is used
If ED does not detect GSM signal, the correlation detector is used
Can achieve performance of the correlation based detector
– Which has much higher computational complexity
Spectrum Sensing Summary IIT Madras
Many methods available

– Properties utilised: Signal energy, Correlation, Cyclostationarity
– Computational complexity and estimation time are important factors
– Searching over a vast frequency range
Focus on robustness (at low SNR) and reliability
Minimize probability of missed detection
– To avoid interference to primary user
Uncertainties regarding measurement
– Noise and interference environment
Strong motivation for Hybrid Detectors
Sensing Criteria (Regulatory aspects)
– Sensing Period
– Detection Sensitivity

Regulatory Constraints IIT Madras
Satisfactory protection of primary user from harmful interference

– Essential for realization of opportunistic spectrum access
– Regulatory constraints
z Sensing Periodicity (Tp)
– Period with which UL user must check for presence of primary user
z Detection Sensitivity
– Signal level at which the UL user must detect primary user reliably
Sensing Period (Tp)

– Max. time (delay) UL user unaware of reappearance of primary user
– Max. duration of harmful interference
– Determines QoS degradation of primary user
– Delay of primary user in accessing channel
– Depends on type of primary user service – delay sensitivity
– Must be set by regulator for each licensed band

Detection Sensitivity IIT Madras
CR network must avoid harmful interference to PU

– SIR of PU must not fall below threshold (Γ )
Threshold (Γ ) depends on
– PU receiver’s robustness to interference
z Example: SIR = 34 dB for analog TV, and SIR = 23 dB for digital TV
– Characteristics of interfering signal
z Power, waveform, continuous vs intermittent, … )
– May influence choice of transmission option chosen by CR
Interference Range of secondary user

– Max distance from Primary user at which harmful interference occurs
Pp L(R ) Pp , Ps , Pb = Power of PU, SU, and background noise + interference

Γ=
Ps L(D ) + Pb
R = Max distance between PU transmitter and receiver
L(d ) = Path loss (Shadowing and Fading) at distance d
Detection Sensitivity IIT Madras
Ref: Ghasemi et al., IEEE Communications Mag,

April 2008
Pp L(R )
Γ=
Ps L(D ) + Pb
Pp , Ps , Pb = Power of PU, SU, and background noise + interferen ce

R = Max distance between PU transmitte r and receiver
L(d ) = Path loss (Shadowing and Fading) at distance d
Threshold to be satisfied even if PU Rx is at edge of coverage

– Provided SU maintains distance D
⇒ SU (CR) must be able to detect PU at distance (R+D)
Pp L(D + R )
Detection Sensitivity γ min =
Pn
Detection Sensitivity (3) IIT Madras
Pp L(R ) Pp L(D + R )
Γ= γ min =
Ps L(D ) + Pb Pn
Detection Sensitivity ( γ min )
– The minimum SNR at which PU signal must be reliably detected
z Example: Probability of detection = 0.99
Regulator must specify the following
– Γ, R, Pp ⇒ γ min can be obtained

Strong dependence between γ min and Ps
As γ
min ↓ ⇒ Ps ↑

Also, if the number of secondary users increases, Ps ↓ or D↑

Uncertainties in Sensing IIT Madras
Channel Uncertainty
Due to fading / shadowing of PU signal
– ⇒ Higher detection sensitivity requirement
– Benefit from cooperative sensing (by multiple CR devices)
– Increasing sensing period Tp will help
– Antenna diversity is a benefit
Noise Uncertainty
Rx noise power level not known a priori → has to be estimated
Can have variations (due to temperature, calibration, … )
Significant impact on performance if energy detector is used
Weak PU signal indistinguishable from noise
– If SNR falls below threshold
Feature detectors (e.g., cyclostationarity-based) are more robust
Uncertainties in Sensing IIT Madras
Channel Uncertainty
Due to fading / shadowing of PU signal
Noise Uncertainty
Aggregate Interference Uncertainty

PU may experience harmful
interference
– If multiple CR networks active
Requires more sensitive detectors
– Detect PU at distance
D′ > ( D + R )
Ref: Ghasemi et al., IEEE Communications Mag, April 2008
Alternative – system level coordination among CR devices

– ⇒ Cooperative sensing

Cooperative Sensing Electrical Engineering
IIT Madras
To alleviate the problem … → Cooperative Sensing
– Independent measurements at different locations / CRs
– Exchange of sensing information among CR nodes
– Diversity gain achieved (handles fading and shadowing)
– Improved probability of detecting PU
z Without increasing sensitivity of each individual SU Rx
– Introduces additional communications overhead
– Requires functionality of “Band Manager” (Fusion Centre)
z Collects information, makes decisions and shares information with all CR nodes
Shadowing is correlated over short distances
– → Cooperation to be done over larger distances (few nodes)
– Different from conventional view of Mesh / Ad Hoc networks (many nodes in close
proximity)

IIT Madras
Multicarrier Techniques in CR

Multiple Access IIT Madras
Code
Code WCDMA,
cdma2000
CDMA TD-SCDMA
802.11b OFDM
Frequency
Frequency Time
802.11 a/g
Time 802.16
802.20?
Code
Code
TDMA FDMA
GSM,
EDGE
Frequency Frequency
Time
Time
OFDM Electrical Engineering
IIT Madras
Effect of Multipath
– CDMA – loss of orthogonality
z More severe if spreading factor is low
– TDMA ⇒ need for complex equalization
z More severe for higher baud rates
– OFDM attractive for high speed data in multipath fading
OFDM – Orthogonal Freq Division Multiplexing (Multicarrier)
– Narrow carriers ⇒ low baud rate ⇒ long symbol duration
– An attractive candidate for broadband wireless
– Efficient digital multicarrier implementation using DFT/IDFT
– Opportunity to do optimized coding and modulation in each carrier
z Maximize capacity utilization based on channel condition
z A active area of research
– Issues: High peak-to-average ratio, sensitivity to frequency & timing errors

OFDM used for WLAN, WWAN, Digital Audio Broadcasting, 4G, …
OFDM ⇔ Multi-carrier Modulation ⇔ Multi-tone Modulation
Multicarrier Techniques IIT Madras
Multicarrier techniques widely used in Cognitive Radio (PHY)

– OFDM, Filterbank-based multicarrier, Multi-resolution filter banks
– Spectrum sensing – determine spectral holes
– Spectrum usage – communication
z Transmit data w/o interfering with Primary user
z In non-overlapping parts of spectrum
Multicarrier techniques – efficient and effective

CR transmission can be TDD or FDD
TDD has inherent advantages for CR
– Tx and Rx in in same band ⇒ knowledge of channel
z Implicit sensing of channel during Rx period (Tx OFF)
802.22 WRAN standard focus on TDD
– OFDM based

OFDM Carriers in Available Spectrum IIT Madras
Spectral Adaptation Waveforms
T
I
M
E
Frequency
Ref: B. Fette, “SDR Technology Implementation for the Cognitive Radio,” General Dynamics

OFDM
– Widely studied and well-understood (based on IFFT / FFT)
– Used for spectral sensing
– Underlying filter is the Rectangular window
z Poor side-lobe suppression
z Significant interference between sub-carriers
– Not suitable for spectral sensing / transmission (non-contiguous bands)

z Acceptable for contiguous bands
Approaches to consider
– Muti-Taper Method (MTM) for spectral estimation
– Filterbank Multi-Carrier
Filterbank-based approaches can overcome spectral leakage problems
– Less used than OFDM

Filterbank view IIT Madras
Raised cosine filtering before FFT

– Reduces side-lobes
Improved freq selectivity
– At expense of lower time selectivity
Filterbank Multicarrier
– Length 6x256=1536,
– 256-channel filterbank
Frequency response of “FFT filter”

φi ( f ) = K sinc 2 (( f − f i ) Ts )
f i = Centre frequency of i th sub - channel
Ts = OFDM symbol period (incl. CP)
Ref: Boroujeny et al., IEEE Communications Mag, April 2008

Multitaper Method (MTM)

– Advanced, non-parametric spectral estimation method
– A set of filters (Slepian 1978, Bell Labs)
z Discrete Prolate Spheroidal Sequences (DPSS)
z Optimal trade-off between
– time selectivity and frequency selectivity
– Combine the output of a family of filters
MTM – five filters of length 2048
– Near-optimal performance in spectral sensing (Haykin, 2005)
– Example: A set of 5 DPSS based filters and their responses
Filterbank Method
– Similar performance to MTM
– Can be used for sensing and for transmission
– Lower computational complexity than MTM

Multicarrier Approach IIT Madras
Orthogonal Frequency Division Multiplexing
Orthogonal Frequency Division Multiplexing with Filterbank
Ref: Wyglinski et al, “Cognitive

Radio Communications”
Academic Press, 2010
FB-MC subcarrier spectrum employing a square-root

raised cosine prototype lowpass filter with a rolloff of 0.25
54
OFDM based CR Electrical Engineering
IIT Madras
OFDM ideally suited for CR

OFDM based CR Electrical Engineering
IIT Madras
OFDM well-suited for CR applications
Ref: Arslan., IEEE Wireless Communications April 2009

IIT Madras
IEEE 802.22
CR-Based Wireless Regional Area
Network (WRAN)

IEEE 802.22 IIT Madras
Project started by IEEE in Nov 2004

Charter: To develop a CR-based WRAN
– PHY and MAC specifications
Transmission in unused TV and guard bands (54 MHz – 862 MHz)
– Very favourable propagation characteristics
– Channel BW 6 MHz (may be 7 MHz / 8 MHz in some countries)
Spectrum sensing for identifying white spaces
– Distributed sensing
z FCC maintained server – info about unused channels (by geographical location
– Localised sensing
z CPE’s perform periodic measurements and send measurements to BTS
z BTS makes decision to use the current channel or any other alternatives
Application scenarios
– Wireless broadband in rural / remote areas
z Performance comparable to today’s DSL technology
– Unlicensed devices ⇒ lower cost and increased affordability

– TV migration : moving from broadcast to cable and satellite
z ⇒ Broadcast TV channels available

Comparison of Networks IIT Madras
WRAN Aspects
Large coverage footprint
– Up to 100 Km
Larger cells than cellular
Leverage two factors
– Higher EIRP
– Attractive propgn characteristics
Ideal for rural /remote services
– Broadband wireless access
Unlicensed devices
Ref: Cordeiro et al., “IEEE 802.22: The First Worldwide

Wireless Standard based on Cognitive Radio,” IEEE, 2005

IEEE 802.22 Specifications IIT Madras
Target specifications
– Spectral efficiency – 0.5 b/s/Hz – 5 b/s/Hz
z Average: 3 b/s/Hz ⇒ 18 Mbps in 6 MHz
z Assuming 12 simultaneous users – 1.5 Mbps (DL) and 384 Kbps (UL)
– Range: 33 Km (extend to 100 Km)

– CPE Tx power 4W EIRP @ CPE
Air interface
– Requirements – Flexibility and quick adaptibility
z Link adaptation based on SINR
z Adapt modulation and Coding option
z Frequency agility
– OFDM(A) based UL and DL

– Transmit Power Control : 30 dB with steps of ± 1 dB
– Channel Bonding – Utilizing more than one TV channel
z System can use larger BW to support higher throughput

IEEE 802.22 MAC IIT Madras
Medium Access Control (MAC)

– Design tailored for Cognitive Radio functionality
z Channel sensing
z Channel classification
– Operating ⇒ currently being used (sensing every 2 sec)
– Back-up ⇒ Cleared for becoming operating channel (sense every 6 sec)
– Candidate ⇒ likely to become back-up channel (every 6 sec, for 30 sec)
– Protected ⇒ incumbent detected through sensing (every 6 sec)
– Disallowed ⇒ due to operational or regulatory issues
– Unclassified ⇒ not yet sensed
z Maintenance of channel information
Status of TV channels from external geo-location database
– List of available TV channels + permissible EIRP
MAC deals with “protection of incumbent” not addressed in traditional systems

ATSC Signal Sensing IIT Madras
Spectrum Shape and Signal Characteristics

– More efficient than Analog in Bandwidth and Power
– VSB spectrum is flat
z noise-like randomized data
– SRRC pulse shaping is used (α=0.115)
– Half Power freqs are 5.381 MHz apart
– Low-Level Pilot Carrier
z At lower band edge
z Adds 0.3 dB to total avg power
Data Structure and Moduln Scheme
– 8 data levels (leads to 3 bits/symbol)
– Has SYNC symbols occurring (4 out of every 832 symbols)
z Can be exploited for signal detection

Detection Techniques IIT Madras
Proposed Approaches :
1. Energy Detection
2. Detection Using Tones
3. Thomson’s Multitaper Method (Haykin, Reed and Thomson)
4. Cyclostationary Detection
Tone Detection
– Extensively used in GSM
– BCCH detection methods
– Robust detection techniques like DTMF

Detection Techniques IIT Madras
Multitaper Method
– Non-parametric sensor
– These methods use tapering
– Windowing causes the bias-variance
– Multi-taper method uses Slepian sequences
z Max-energy concentration property
– The data sequence is multiplied with these

N −1 C0 = N W
Xk(f ) = ∑ x (n )v exp( − 2π fn )
(k )
n K ≤ ⎣2 N W ⎦
n=0 C0 Sidelobe
(k )
vn are a set of Slepian Tapers energy (dB)
4 − 95
− 149
Max. Energy Concentration in ( f − W / 2, f + W / 2)
6
–
8 − 203
– W - Resolution of the estimator 10 − 257

Multi-Taper Method IIT Madras
Spectral Concentration Problem

Find finite-duration sequence with Max Spectral Concentration in given band
N
U ( f ) = ∑ wn e − 2πft
n =1
Spectral Concentration is defined as

W
∫
2
U ( f ) df
λ ( N ,W ) = −W
0.5
∫
2
U ( f ) df
− 0.5
A finite sequence - cannot give finite bandwidth spectrum

0 < λ ( N ,W ) < 1
Formulated as an eigenvector problem
Eigenvectors are called Discrete Prolate Spheroidal Sequences

Plots of First Four DPSS IIT Madras
First four Slepian DPSS seqs
First four Slepian DPSS seqs

C0=6.5, K=11 tapers, N=2200
20 sets of data
10.72 Msymbols per sec
MTM IIT Madras
Multitaper Method
– Check for Line spectra within ( f −W / 2, f +W / 2)
– Bias-Variance problem is replaced with Bias-Resolution problem
– C0 = NW Time-BW product limits number of tapers used

– λk are the eigenvalues associated with the kth eigenspectrum
K ≤ ⎣2 NW ⎦
(1 − λk ) give a measure of leakage in the kth eigenspectrum
– A natural spectral estimator is
K −1
∑ λk X k ( f )
2
Sˆ ( f ) = k =0
K −1
∑
k =0
λk

IIT Madras
Interference Mitigation - Relays

Indoor Personal Relay (IPeR) IIT Madras
Conventional Femto basestation not feasible in Indian context

– Lack of wireline broadband for backhaul ⇒ “wireless femto”
Relay has directional antenna or multi-antenna link to eNodeB
– High-SINR link
IPeR is user-deployed
Tx power controlled by BS
– Interference must be minimized
Potential for high bit rate indoors
– x2, x3 throughput for indoor users
Near-Transparent L1 relay
– With selective forwarding
– Low latency
Cost, complexity, and power similar to terminal
Interference Case I IIT Madras

Interference Case II IIT Madras

Interference Case III IIT Madras

Managing Interference IIT Madras
Relays use the same spectrum as macro-cell in uplink and downlink

– Typically a relay serves only a few users in a small area
Interference management can play an important role in ensuring that
– Gains from deployment of relays are maximized
– Spectrum is utilized efficiently
Current solutions orthogonalize interference
– static (or semi-static) partitioning of resources between eNB and RNs
– Potentially under-utilization of resources
Cognitive Interference Management (CIM) – a dynamic technique
– “Cognitive” ⇒ adapts or learns from environment
– Completely avoiding interference from relays
– Improving spectrum re-use

Cognitive Interference Management IIT Madras
Objective:
– Dynamically allocate resources
– Interfering links become orthogonal
The cognitive framework comprises two elements
– UE Classification
z To identify interfering links
– Scheduling of transmissions such that
z eNB-UE and RN-UE links that do not interfere can use same resources
z eNB-UE and RN-UE links that interfere use orthogonal resources
Allows any part of frame to be used for eNB ↔ RN and RN ↔ UE
– RNs do not have to transmit and receive at the same time
– Interference constraints are satisfied
RN power control used - additional tool to change interference profile
UE Classification IIT Madras
Identify interfering links, eNB classifies UEs w.r.t to each RN

– Relay-cell UE - Served by the RN
– Victim UE – Sees significantly strong interference from the RN
– Safe UE – All other UEs (not affected by interference from the RN)
Victim or Safe UE is being served by the eNB or another RN

Example: Classification IIT Madras
RN1 RN2
UE1 R S
UE2 V S
UE3 S V
UE4 S S
UE5 S R
R – Relay-cell UE
S – Safe UE
V – Victim UE

IIT Madras
CR Testbed

CR Testbeds IIT Madras
Many universities have implemented CR testbeds
– VA Tech, Berkeley, U Kansas, Rutgers-Winlab, Georgia Tech, …
In India, CDAC Thiruvananthapuram developing CR Testbed
– Focus on spectrum sensing
– Algorithms developed by IISc Bangalore
Testbed needed for algorithm validation, prototype development

VA Tech CORNET IIT Madras
VA Tech CORNET (Cognitive Radio Network) as a case study
– Focus on SDR and CR capability
z Flexibility in PHY and MAC layers - Enable processing intensive research
– 48 nodes
– Each node has two main componets
z General purpose computing platform (Intel Xeon processor based server)
z Flexible RF front-end
– RF Component
z Ettus Research USRP2 (Universal Software Radio Peripheral)
– Supports rapid design and implementation of SDR systems
– Motherboard – VirtexSpartan3 FPGA (high-speed signal processing)
– 14-bit, 100 Msps Analog → Digital,16-bit, 400 Msps Digital → Analog
z Daughter board – Motorola RFIC4
– Continuous tuning in 100 MHz – 4 GHz, Instantaneous BW 10 KHz – 20 MHz
– Gigabit Ethernet connection between USRP2 and host processor


IIT Madras
IITM Testbed

Generic Wireless Transceiver Electrical Engineering
IIT Madras
Baseband LPF RF up-

DAC
processing converter
Baseband ADC CSF RF down-

processing converter
Baseband
processor

Wireless Transceiver Testbed Electrical Engineering
IIT Madras
Transmitter Board Receiver Board

FPGA FPGA
RJ45 Analog RJ45
DSPs Interface DSPs
Phy layer BB processor Phy layer BB processor Ethernet

Ethernet-
backend
backend
MAC layer MAC layer

processor processor

Baseband Processing Electrical Engineering
IIT Madras

Remote Radio Head-end Electrical Engineering
IIT Madras

Setup for 4x4 Antenna Configuration IIT Madras

Summary IIT Madras
A broad overview of Cognitive Radio

A paradigm shift in wireless communications
Potential of significant increase in spectrum availability
– Opportunistic access
Spectrum sensing – a key element in CR
Cooperative sensing is attractive
IEEE 802.22 standard – an interesting opportunity
CR techniques in 4G Indoor Personal relays
Overall, CR is an exciting field

IIT Madras
My best wishes
to all participants of
the IWCR 2010 Workshop
…
Thank You !

David Koilpillai Profile IIT Madras
Education
B.Tech, IIT Madras, MS, PhD Caltech, USA
Work Experience
IIT Madras (2002 – present)
Professor, TeNeT Group, EE Department
z
CEWiT – Chief Scientist (Jan 2007 – July 2007

Co-Chair, IIT Hyderabad Task Force (June 2008 – Dec 2009)
Ericsson Inc, USA (1990-2002)
z Director, Advanced Technologies, Research and Patents
(R&D team of 75 engineers, annual budget US $20 Million)
Professional
– Areas of expertise: Cellular, wireless systems, DSP
– 32 Issued US patents
– Publications: 11 Journal, 45 Conference
– Research Interests: DSP applications in Wireless
– Ericsson Inventor of Year Award 1999
– Fellow, Indian National Academy of Engineering

Koilpillai CR Keynote

Загружено:

Сведения о документе

Исходное описание:

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Koilpillai CR Keynote

Загружено:

Авторское право:

Доступные форматы

Electrical Engineering

Signal Processing Aspects in Cognitive Radio

International Workshop on Cognitive Radio (IWCR 2010)

Koilpillai / Dec 2010 / Cognitive Radio 1

 Wireless today – SDR, Wireless Broadband, …

Koilpillai / Dec 2010 / Cognitive Radio 2

1G (AMPS, NMT, TACS, …) 1981

2G (GSM, IS-54, PDC, IS-95) 1991 - 95

2.5G (GPRS, cdmaOne) 1999 - 00

3G (WCDMA, EDGE, cdma2000) 2002 - 03

3.5G (HSPA, 1xEV, IEEE 802.16e) 2003 - present

4G (LTE, IEEE 802.16m, …) 2010+

3.5G/4G - Focus on high data rates, spectral efficiency

 Hierarchy of terrestrial networks

Koilpillai / Dec 2010 / Cognitive Radio 6

Ref: Kori (Alcatel Lucent) –

 Compelling case for

Koilpillai / Dec 2010 / Cognitive Radio 7

 ITU initiative (started 06/2003)

 Spectrum auctions (3G+BWA) held during April-June 2010

 Dense Urban (Case: Mumbai)

– ~3733 households per sq km

– ~ 50% wireless internet subscribers

 Typical scenarios evaluated by Indian operators

Koilpillai / Dec 2010 / Cognitive Radio 10

Software Defined Radio (SDR)

Koilpillai / Dec 2010 / Cognitive Radio 11

Source: Prasad et al. IEEE Comm Magazine, April 2008

J. Mitola, “The software radio architecture”

Koilpillai / Dec 2010 / Cognitive Radio 13

 Software radio approach (cost effective and computationally efficient)

– Narrowband systems – requiring large number of transceivers

RF stage IF stage Baseband

 SDR approach (multi-channel receiver)

RF stage IF stage High speed DSP-based Baseband

Koilpillai / Dec 2010 / Cognitive Radio 14

 Very stringent A/D requirements

 SDR ⇒ Flexible and efficient receiver and transmitter architectures

Koilpillai / Dec 2010 / Cognitive Radio 17

 Many technical challenges have been solved

SDR → Cognitive Radio

Koilpillai / Dec 2010 / Cognitive Radio 19

 Increasing demand for radio spectrum

 Cognitive Radio techniques – much broader than DSA

Koilpillai / Dec 2010 / Cognitive Radio 20

Increasing demand for spectrum

 CR: Opportunistic Unlicensed Access

Koilpillai / Dec 2010 / Cognitive Radio 23

SDR → Cognitive Radio

Koilpillai / Dec 2010 / Cognitive Radio 25

Koilpillai / Dec 2010 / Cognitive Radio 26

Koilpillai / Dec 2010 / Cognitive Radio 27

 Alternative – Energy Detector

– Uncertainty of noise floor will degrade performance - Especially at low SNR

 ED can be effectively combined with more robust detectors – “Hybrid Detectors”

Koilpillai / Dec 2010 / Cognitive Radio 28

 Binary hypothesis testing problem

H 0 : y [n] = w [n] ⇒ Primary User absent

 If N large, invoke CLT

 Primary user signal → GSM signal

 Performance plots for Energy Detector

 ED is a strong candidate for first stage of Hybrid detector

Koilpillai / Dec 2010 / Cognitive Radio 31

Wireless today – SDR, Wireless Broadband, …

Hierarchy of terrestrial networks

Compelling case for

ITU initiative (started 06/2003)

Spectrum auctions (3G+BWA) held during April-June 2010

Dense Urban (Case: Mumbai)

Typical scenarios evaluated by Indian operators

Software radio approach (cost effective and computationally efficient)

SDR approach (multi-channel receiver)

Very stringent A/D requirements

SDR ⇒ Flexible and efficient receiver and transmitter architectures

Many technical challenges have been solved

Increasing demand for radio spectrum

Cognitive Radio techniques – much broader than DSA

CR: Opportunistic Unlicensed Access

Alternative – Energy Detector

ED can be effectively combined with more robust detectors – “Hybrid Detectors”

Binary hypothesis testing problem

If N large, invoke CLT

Primary user signal → GSM signal

Performance plots for Energy Detector

ED is a strong candidate for first stage of Hybrid detector

Robustness of energy detector enhanced if longer sensing period is used

Well-suited for signals like GSM

Correlation provides processing gain

Cyclostationary Property – Digital modulated signals

Focus on GMSK signal (BT=0.3)

Exploiting multiple cyclic frequencies improves performance

A robust detector for GSM signals (GMSK)

Many methods available

Satisfactory protection of primary user from harmful interference

Sensing Period (Tp)

CR network must avoid harmful interference to PU

Interference Range of secondary user

Threshold to be satisfied even if PU Rx is at edge of coverage

Also, if the number of secondary users increases, Ps ↓ or D↑

Alternative – system level coordination among CR devices

Multicarrier techniques widely used in Cognitive Radio (PHY)

Multicarrier techniques – efficient and effective

Raised cosine filtering before FFT

Frequency response of “FFT filter”

Multitaper Method (MTM)

Orthogonal Frequency Division Multiplexing

Orthogonal Frequency Division Multiplexing with Filterbank

OFDM ideally suited for CR

OFDM well-suited for CR applications

Project started by IEEE in Nov 2004